We used data from women who participated in Project Viva for this analysis. Project Viva is a prospective cohort study that recruited pregnant women carrying a singleton pregnancy during their initial obstetric care visit at Atrius Harvard Vanguard Medical Associates in eastern Massachusetts between 1999 and 2002 . Study exclusion criteria included multiple gestation, inability to answer questions in English, gestational age ≥ 22 weeks at recruitment, and plans to move away from the study area before delivery . There were 2128 births to 2100 mothers (28 women enrolled with more than one pregnancy). Project Viva attempted to follow-up all participants thereafter, most recently at a mid-life follow-up study visit at a mean (SD) of 18.2 (0.6) years after study enrollment, conducted 2017–2021. 
For this analysis, we included Project Viva women who reported information on lifetime breastfeeding duration and were tested for handgrip strength at the midlife visit. A Project Viva research assistant administered pre-test questions. If the participant 1) had hand or wrist surgery on both hands within the last three months or 2) was unable to hold the dynamometer with both hands (e.g., missing arms, hands, or thumbs on both hands; paralysis of both hands) then the research assistant did not measure handgrip strength. If the participant had hand or wrist surgery on one hand within the last 3 months or was unable to hold the dynamometer with one hand, they performed the handgrip strength exam only using the opposite hand. Of the 2100 women enrolled in Project Viva, 676 provided handgrip strength data at the midlife visit and 631 provided lifetime breastfeeding duration data. The participants included in this analysis had similar socioeconomic statuses and race/ethnicity proportions as those without mid-life follow-up data, however there was a slightly larger percentage of college graduates in those included versus excluded (75% vs. 60%) (Additional file 2: Table S1).
We defined the primary exposure variable as total lifetime breastfeeding duration across all pregnancies, measured continuously per 3-month increment. At the midlife visit, women reported information about all lifetime pregnancies (not limited to the Project Viva index pregnancy) via questionnaire, including year of pregnancy (from which we calculated age at first pregnancy) and duration of breastfeeding for each pregnancy that resulted in a live birth. We calculated total lifetime breastfeeding duration by adding the total duration for each reported pregnancy. We also calculated total lifetime breastfeeding duration ranked into quartiles (with quartile 1 as the reference category), dichotomous ever breastfed (yes vs. no) and average breastfeeding duration per live birth (per 3-month increment).
The outcome variable was midlife handgrip strength. Trained research assistants measured handgrip strength at the midlife visit in kilograms using a Jamar dynamometer. The Jamar dynamometer is a validated instrument for measuring handgrip strength and serves as a reference standard across clinical and epidemiological studies [16, 17]. Using the American Society of Hand Therapists protocol, participants were instructed to squeeze the dynamometer with maximal effort for three separate attempts per hand with thirty seconds of rest in between to avoid muscle fatigue . We used the average measurement of the three attempts separately for the dominant and non-dominant hand in our analyses and also calculated the average of all 6 measures.
We considered a priori covariates that may be associated with both the exposure and outcome in our analysis, based on review of the literature and using directed acyclic graphs. Women reported annual household income via a self-administered questionnaire . At study enrollment, women reported, highest educational attainment, and marital status via interview. At study enrollment, research assistants also asked mothers: “Which of the following best describes your race or ethnicity?” Mothers had a choice of one or more of the following mutually exclusive racial/ethnic groups: Hispanic or Latina, White or Caucasian, Black or African American, Asian or Pacific Islander, American Indian or Alaskan Native, and other (please specify). For the participants who chose “other” race/ethnicity, we compared the specified responses to the U.S. census definition for the other five races and ethnicities and reclassified them where appropriate. If a participant chose more than one racial/ethnic group, we coded them as “more than 1 race/ethnicity”. We chose to adjust for race/ethnicity because we consider it to be a social construct that may affect breastfeeding duration through a variety of mechanisms; there are also differences in body composition proportions across race/ethnicity . We also chose to adjust for age at first pregnancy because it is inversely related to lifetime breastfeeding duration and could potentially confound the association between lifetime breastfeeding duration and midlife handgrip strength . At enrollment, pregnant women reported their diet during the index pregnancy using a validated food frequency questionnaire [19, 20]. The food frequency questionnaire used in Project Viva was modified for pregnant adults from the Willett FFQ used in the Nurses’ Health Study and other large cohort studies [15, 21, 22]. For our analyses, we calculated Alternate Healthy Eating Index, slightly modified for pregnancy (AHEI-P)  as a measure of overall diet quality. At enrollment women also reported pre-pregnancy physical activity using a questionnaire modified from the Physical Activity Scale for the Elderly (PASE) . Women were asked to recall their weekly activity over the year before pregnancy and to report average hours per week of activity . Total physical activity was treated as a continuous covariate (hours/week). At the midlife visit, women reported whether they had ever smoked cigarettes. Older age is inversely associated with handgrip strength measurements . Therefore, age at handgrip strength assessment was identified as a precision covariate for handgrip strength because it accounts for variation in handgrip strength.
We considered pre-pregnancy weight as a covariate as previous studies have found that BMI has a positive association with handgrip strength [25, 26]. We could not adjust for weight prior to all pregnancies, as we did not collect this variable. We did collect estimated weight status at 10 years of age, reported at study enrollment, but after adjusting for it in a sensitivity analysis, it did not meaningfully change the results, so we did not include this variable in our models.
Lifetime parity was found to be moderately associated with increased lifetime breastfeeding duration (Spearman r = 0.33, p < 0.0001). We accounted for parity by using average breastfeeding duration per live birth.
In our primary analysis, we estimated associations of lifetime breastfeeding duration in months (continuous, reported per 3-month increment to make estimates meaningful in magnitude) with midlife handgrip strength in kilograms using unadjusted (Model 1) and multivariable adjusted linear regression. We tested for normality of the exposure distribution and decided to use the non-transformed lifetime breastfeeding duration for ease of interpretation as both the log2 transformed and nontransformed lifetime breastfeeding duration yielded models with normally distributed residuals. We adjusted for race/ethnicity, education, marital status, ever smoking, household income at enrollment, and mother’s age at first pregnancy (Model 2). Model 3 was also adjusted for age at handgrip strength measurement due to the inverse relationship between age and handgrip strength . Finally, in sensitivity analyses, we additionally adjusted for diet (AHEI-P , units) and pre-pregnancy physical activity (hours/week) as both can impact maternal body composition and therefore impact breastfeeding outcomes [26,27,28,29]. However, because we had information on pre-pregnancy diet and physical activity only for the index pregnancy, we restricted this sensitivity analysis to women who enrolled in Project Viva at their first pregnancy (Model 4). We then repeated all regression models using quartiles of lifetime breastfeeding duration in months as the exposure and compared differences in mean handgrip strength using the first quartile as the reference.
To assess for racial/ethnic differences in exposure-outcome associations, we added an interaction term between lifetime breastfeeding duration and race/ethnicity and also examined results in models stratified by race/ethnicity. We considered evidence for significant interaction if the interaction p value was < 0.15.
We repeated Models 1 to 4 using the exposure average breastfeeding duration per live birth as a standardized indicator of breastfeeding duration irrespective of parity.
We performed all analyses using SAS version 9.4 (Cary, NC). Due to the small number of missing covariate values, we did not use multiple imputation and allowed the sample size to decrease slightly across multivariable models.